Sign in / up

back to article At what point do servers become HPC beasts?

From time to time, The Register commissions its own "barometer surveys", to gauge the impact of technologies on our readers' working lives Today, we launch a new barometer survey, to track high-end servers and HPC, or high performance computers, the name by which low- and mid-range supercomputers go by these days. We will run …

COMMENTS

House rules Send corrections
This topic is closed for new posts.
  1. Henry Wertz 1 Gold badge

    Is this an opportunity?

    "To many, Linux clusters are still a black art - is this an opportunity for Johny-come-lately Microsoft. "

    I doubt it.

    1) They've had HPC software technically available for years, with very few takers.

    2) Who would really want to pay an extra cost per box in a cluster just to run Windows? No-one's going to be running Office, Quicken,etc. on it (no "legacy software" advantage)... they aren't going to be throwing random hardware onto it, negating the (I think rather overblown) driver advantage. And they are probably headless, usually HPC systems have a seperate visualization frontend (so no advantage there.). I could see someone using Direct3D instead of OpenGL for that system I suppose (although I wouldn't.)

    It's quite true though, servers have at least as fast CPUs as desktops, with superior memory bandwidth, I/O bandwidth, and more CPUs per box then desktops tend to have. This would make them good candidates for an HPC cluster. The main downside traditionally was weak on-board video with either a low-profile or no AGP slot (I don't know if this trend continues with PCI Express servers...) Of course for HPC use, the boxes will probably be set up once then run headless anyway. The other tradeoff is that servers are more costly, memory is more costly (registered ECC versus plain ol' RAM?) and so on.

    0 0
  2. Robert Hill
    Boffin

    And we didn't even get a chance to sign it...

    All that typing, and we couldn't even sign our input. Blah.

    Mine's the one that states the three biggest trends as:

    1) GPUs as commodity numeric processors have finally put deskside supercomputers in our reach, and will change the way businesses look at data - and how much data they look at.

    2) Stream processing is finally moving out of academia and into the business world, and once business figures out what it IS then we are in for a whole host of new applications...the same way "Harvest" revolutionized codebreaking for the NSA.

    3) The recent coming of real (and practical) 3D graphical displays on desktops and laptops will create a HUGE demand for HPC for 3D content creation and manipulation - probably larger than anyone currently thinks.

    Anyway, you read them here first...lol

    0 0
  3. CoonDoggy

    ECC?

    > The other tradeoff is that servers are more costly, memory is more costly (registered ECC versus plain ol' RAM?)

    When I can get 4G of ECC ram for $100, I see no reason to run a HPC cluster without it - single bit errors are way more serious when you're crunching a TB of data for 2 days.

    0 0
  4. Robert Hill
    Boffin

    More "Final Thoughts"...if I may

    To add to my thoughts in the survey, I would like to point out that one of the limitations on HPC growth in general business applications is the relative paucity of business executives that thrive on data and are quant driven. Far _fewer_ are the number of executives that can actually balance the use of sophisticated quant data derived by HPC, and still have the political saavy and business culture knowledge necessary to successfully create new areas of business opportunity powered by HPC. This is and will be addressed in business schools, especially with cheaper HPC resources being available to those in B-school. But it will take years for those with that knowledge to permeate our business culture...as usual, the change management takes longer than the IT implementation...

    0 0
This topic is closed for new posts.

Other stories you might like